The speciﬁc-word frequency effect in speech production...

The specific-word frequency effect in speech production:Evidence from Spanish and French

Fernando CuetosUniversidad de Oviedo, Oviedo, Spain

Patrick BoninLEAD/CNRS, University of Bourgogne, Dijon, France

Jose Ramon AlamedaUniversidad de Huelva, Huelva, Spain

Alfonso CaramazzaHarvard University, Cambridge, MA, USA, and Center for Mind/Brain Sciences, University of Trento, Trento, Italy

The role of word frequency in lexical access during the production of homophones remains un-resolved. In the current study, we address whether specific-word (the frequency of occurrence ofthe word “nun”) or homophone frequency (the summed frequency of words with the pronunciation/nLn/) determines the production latencies of homophones. In Experiments 1a, 2a, and 3a, partici-pants named pictures of high-frequency (e.g., “banco”–a bank: financial institution) and low-frequency (e.g., “banco”–park bench) Spanish (Experiments 1a and 2a) or French (Experiment 3a)homophones and control pictures of nonhomophone words matched in frequency with each of thetwo uses of the homophones. The naming latencies for low-frequency homophones were longerthan those for high-frequency homophones. Furthermore, the naming latencies for homophoneswere indistinguishable from those for nonhomophone controls matched in specific-word frequency.In Experiments 1b, 2b, and 3b, the participants performed either object decision or picture–wordmatching tasks with the stimuli used in the corresponding Experiments 1a, 2a, and 3a. There wereno reliable differences between high- and low-frequency homophones. The findings support thehypothesis that specific-word and not homophone frequency determines lexical access in speechproduction.

Keywords: Lexical access; Picture naming; Homophones.

The influence of homophones on word productionwas investigated for the first time by Dell (1990).Using an error induction technique, Dell showedthat the phonological errors produced by the

participants were determined not by the specificfrequency of the target word, but rather by thecombined frequency of the word and its homo-phones. Several years later, in an influential

Correspondence should be addressed to Fernando Cuetos, Facultad de Psicologıa, Universidad de Oviedo, Plaza Feijoo, s/n,33003 Oviedo, Spain. E-mail: [email protected]

This research was supported by Grant MEC–06–SEJ2006–06712 from the Spanish Government to F.C. and NationalInstitutes of Health (NIH) Grant DC04542 to A.C. We thank Petra Pajtas for editorial assistance.

750 # 2009 The Experimental Psychology Society

http://www.psypress.com/qjep DOI:10.1080/17470210903121663

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY

2010, 63 (4), 750–771

Downloaded By: [Universite Rene Descartes Paris 5] At: 14:03 12 August 2010

article, Jescheniak and Levelt (1994), operatingwith a model of lexical access that assumes thatthere are two lexical layers between a word’smeaning and its phonological content (i.e., thelemma and the lexeme levels, respectively), madetwo important assertions. First, homophones(words that have the same pronunciation butdifferent meanings) share a lexical representationat the lexeme level (morphemic). Second, theword frequency effect in speech production islocated at the level of the shared lexeme represen-tation. These two assertions were based primarilyon the findings of an experiment in whichDutch–English bilingual participants had totranslate English words into Dutch. The exper-iment focused on the translation times for thehomophones, which, because Dutch has a trans-parent orthography, were also homographs (e.g.,the Dutch word “bos”, which has the meanings“bunch” and “forest”). There were three exper-imental conditions: (a) a condition in which thewords were low-frequency homophones (e.g., thebunch sense of bos) with high-frequency homo-phone equivalents (e.g., the forest sense of bos);(b) nonhomophonous words of the same frequencyas the low-frequency homophone (in this case,similar in frequency to the bunch sense of bos);and (c) nonhomophonous words similar infrequency to the total frequency of the homophone(the sum of the frequencies of the bunch and forestsenses of bos). The authors found that the trans-lation times for the homophones were shorterthan those for controls of the same specific-wordfrequency and similar to those for the homophonefrequency controls. Jescheniak and Levelt inter-preted these findings as suggesting that homo-phones share a lexeme representation and thatthe frequency effect is located at the level of thelexeme. In other words, given the assumptionthat homophones share a lexeme representation,it follows that low-frequency homophonesshould inherit the frequency of their homophonemates (see Figure 1).

Somewhat similar results were obtained byJescheniak, Meyer, and Levelt (2003) in anumber of experiments using the same task andprocedure but involving different languages. In

one experiment, the participants had to translatewords from English to Dutch and in anotherexperiment from English to German.

Using a different methodology, Ferreira andGriffin (2003) also obtained evidence thatappeared to support the hypothesis that homo-phones have a common lexical representation.The technique these researchers used was therapid serial visual presentation (RSVP) paradigm.This involves presenting a sentence, word byword, at a rate of 275 ms per word. On criticaltrials, a picture was presented instead of a word,and the participant’s task was to name thepicture. The relationship between the picture andthe expected word was then varied. In semanticallyrelated trials, the relationship between theexpected word and the picture was purely seman-tic—for example, “the woman went to theconvent to become a . . . (expected word ‘nun’)”,followed by the picture “priest”. In contrast, insemantic–homophonic trials, the picture wassemantically related to a homophone of theexpected word—for example, “I thought thatthere would still be some cookies left, but therewere . . . (expected word ‘none’)”, followed by thepicture “priest”. The results showed that the homo-phones of the words that were semantically relatedto the pictures produced interference in naming(e.g., “none” produces interference with “priest”,which is semantically related to the homophone“nun”). (See also Cutting & Ferreira, 1999, forfurther evidence using a picture–word interferencetask.) However, the locus of this effect within theproduction process is not clear.

All of these studies are consistent with thehypothesis that homophones, despite havingdistinct representations at the lemma level,share the same representation at the level of thelexeme (Cutting & Ferreira, 1999; Ferreira &Griffin, 2003). This hypothesis also receivessome support from studies of brain-damagedpatients. For example, Biedermann, Blanken, andNickels (2002) reported the case of an anomicGerman patient, M.W., for whom a word-naming treatment improved naming not just onthe trained words, but also on the homophonesof those words. Similar results were obtained

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (4) 751

SPECIFIC-WORD FREQUENCY


with an English-speaking patient both for homo-graphic homophones (Biedermann & Nickels,2008b) and for heterographic homophones(Biedermann & Nickels, 2008a; but see alsoMiozzo, Jacobs, & Singer, 2004). But here, too,the precise locus of this effect is not clear.

However, other researchers (Bonin & Fayol,2002; Caramazza, Costa, Miozzo, & Bi, 2001;Shatzman & Schiller, 2004) failed to find a homo-phone frequency effect. Furthermore, Caramazzaet al. (2001) found instead that specific-word andnot homophone frequency determines the pro-duction latencies of homophones. In their study,Caramazza et al. used a picture-naming task inwhich three types of stimuli were compared: (a)pictures with homophonic names (e.g., “nun”,which is homophonous with “none”), (b) pictureswhose names were similar in frequency to thespecific-word frequency of the pictured homo-phone (e.g., “owl ”, which has the same frequencyas “nun”), and (c) pictures with names similarin frequency to the cumulative frequency ofthe homophone (e.g., “tooth”, which has thesame frequency as the sum of the frequencies of“nun” and “none”). They found that the naming

latencies of the homophones were similar tothose of the specific-word frequency controls andslower than those of the homophone frequencycontrols. In other words, the time requiredto name “nun” was similar to the time requiredto name “owl ” and different from that needed toname “tooth”.

To confirm that the relatively slow responsetimes for homophones were not due to particulardifficulties in recognizing the pictures correspond-ing to these words, thereby masking an effect ofhomophone frequency, Caramazza et al. (2001)carried out a control picture-naming experimentwith the same pictures in Italian. Since the homo-phone stimuli in English were not homophones inItalian, the participants in the control experimentshould have performed poorly in naming thehomophone pictures if these were particularlyhard to recognize. However, the Italian parti-cipants named the homophone pictures as fastas the low-frequency specific-word pictures.Furthermore, Caramazza et al. replicated thespecific-word frequency effect in picture namingin an experiment withMandarin Chinese speakers.They found that homophone-naming times were

Figure 1. Common versus independent lexical representation models.

752 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (4)

CUETOS ET AL.


similar to those of specific-word frequency controlsand significantly different from those of homo-phone frequency controls. Finally, these research-ers carried out a Spanish–English translationexperiment similar to the translation experimentreported by Jescheniak and Levelt (1994) andagain found that the production latencies forhomophones were similar to those of specific-word frequency controls and significantly slowerthan those of homophone frequency controls.

Bonin and Fayol (2002) used a picture-namingexperiment in which participants were asked toname both members of a homophone pair on sep-arate occasions. For example, they were asked toname both a picture of a glass (the name ofwhich in French, “verre”, is of high frequencyand is pronounced /vEr/),and one of a worm(“ver” in French), which are pronounced thesame way. In other words, if homophones inheritthe frequency of their homophone mates, thereshould be no difference in the naming of the twopictures (e.g., “worm” and “glass”) of a homophone.However, Bonin and Fayol found that the picturesof the high-frequency homophone members (e.g.,“glass”) were named faster than the pictures ofthe low-frequency homophone members (e.g.,“worm”). These findings cannot be ascribed togreater difficulty recognizing the low-frequencymember of a homophone pair since, in a semanticcategorization task in which participants classifiedthe pictures as “natural kind” or “artefact”, partici-pants classified the low-frequency homophonesfaster than the high-frequency homophones.However, as stressed by Biedermann and Nickels(2008a), one potential weakness of the Bonin andFayol (2002) study was the lack of a low-frequencynonhomophone control condition matched to thelow-frequency homophone condition.

Using a similar methodology, Shatzman andSchiller (2004) found that naming latencies wereslower for low-frequency homophones than forhigh-frequency homophones. However, Shatzmanand Schiller (2004) considered that these findingscould be due to differences related to the picturecomplexity, since participants were also slowerwith the low-frequency homophones in a pictureverification task. One possible alternative is that

the differences could have been due to nameagreement since the low-frequency homophoneshad a lower name agreement level than the high-frequency ones.

Miozzo and Caramazza (2005) investigatedword frequency effects in homophones by exami-ning the magnitude of the interference theycreated when presented as distractors during anaming task. Specifically, they found that a homo-phone’s specific frequency was a better predictor ofits effect on picture naming than was its combinedfrequency. In their Experiment 1, they used low-frequency homophones with a high cumulativefrequency (e.g., “brake”) and two controls, oneof similar specific frequency (e.g., “chord ”) andother of similar cumulative frequency (e.g.,“offer”). The results showed that naming latenciesto the pictures in the homophone contexts weresimilar to those of the low-frequency controls.A second experiment used the higher frequencycounterpart of the homophones used in Experiment1 (e.g., “break”), but kept the same control conditions.In this experiment, the interference effects of thehigh-frequency homophones were similar to thoseobserved with the high-frequency controls.

In a more recent study, Gahl (2008) investi-gated the homophone frequency effect by consid-ering the word durations of homophones. Thestudy took advantage of the observation thatfrequent words tend to shorten and that, therefore,if low-frequency homophones inherit the fre-quency advantage of their high-frequency mates,the word durations of the low- and the high-frequency members of a homophone pair shouldbe the same. Gahl (2008) found instead that thehigh-frequency member of homophone pairs(e.g., time) have shorter durations than theirlow-frequency homophones (e.g., thyme). Inother words, low-frequency words do not inheritthe frequency advantage of their high-frequencyhomophones.

Jescheniak et al. (2003; but see Caramazza, Bi,Costa, & Miozzo, 2004) have pointed out thatthere are several potentially important differencesbetween the studies that show a homophone fre-quency effect and those that have failed to findsuch an effect. First, whereas Jescheniak and




Levelt (1994) used homographic homophones intheir experiment, Caramazza et al. (2001) andBonin and Fayol (2002) used heterographic homo-phones.1 They also noted that the control taskused by Caramazza et al. (2001), to assess the poss-ible contribution of differences in the ease withwhich the pictures of homophone and controlwords are recognized, was different from the task(semantic classification) used by Jescheniak andLevelt. Finally, they point out that in the Boninand Fayol study, no controls were used to assesswhether homophone or specific-word frequencyis the better predictor of homophone naminglatencies (see Biedermann & Nickels, 2008a, fora similar criticism).

In the current study, we used a picture-namingtask with homophones to assess whether thereason why Caramazza et al. (2001) and Boninand Fayol (2002) failed to find a homophone fre-quency effect was related to the methodologicaldifferences noted by Jescheniak et al. (2003). Inparticular, we tested whether the differences inthe results reflect the use of languages with trans-parent (Dutch—where a homophone frequencyeffect was found) versus opaque orthographies(English, Mandarin Chinese, and French—where no homophone effect was found). In twoexperiments, we tested speakers of Spanish in apicture-naming task using homographic homo-phones and both specific-word and homophonefrequency controls. Spanish orthography is quitetransparent, and there are many words that havethe same pronunciation and the same spelling,but completely different meanings (for example,“banco” means bank or bench; “vela” means candleor sail, etc.). Furthermore, we carried out controlexperiments that used a semantic classificationtask to assess the possible effects of differences inpicture recognition in the naming task. In a thirdexperiment, we tested French speakers. AlthoughFrench is like English, in that homophones canbe heterographic—for example, “ver” (worm) and“verre” (glass) are both pronounced /vEr/—its

phonology in reading is transparent in the sameway as Dutch and Spanish, since specific ortho-graphic patterns in French are associated withonly one pronunciation. We also introduced acontrol, for specific-word frequency, that was notincluded in the Bonin and Fayol (2002) study.Finally, in the experiments reported here, we alsocontrolled for other variables that could affectnaming times, such as name agreement (seeShatzman & Schiller, 2004) and visual complexity.

EXPERIMENT 1A: PICTURENAMINGIN SPANISH

The aim of this experiment was to determinewhether naming latencies of homophones arebetter predicted by homophone frequency orspecific-word frequency. We used homographichomophonous words such as “banco”, which hasa cumulative frequency of 50 appearances permillion, and specific-word frequencies of 30 and13 for the senses bank and bench, respectively(the remaining seven uses are for other homo-phones such as fish bank or data bank). If thehomophone frequency inheritance hypothesis putforward by Jescheniack and Levelt (1994) iscorrect, there should be no difference in thenaming times for the two homophones (“bank”and “bench”), since the two words would inheriteach other’s frequency, thus resulting in the sameeffective frequency. However, if the specific-wordfrequency hypothesis proposed by Caramazzaet al. (2001) is correct, each different use of“banco” should be associated with a response timethat is specific to its own frequency, independentlyof the frequencies of its homophone mates. Thus,the bench sense of “banco” should have a responsetime similar to that for “cuna” (“cradle”, with a fre-quency of 14), and the bank sense of “banco”should have a response time similar to “chaqueta”(“jacket”, with a frequency of 31).

1 Caramazza et al. (2001) used both homographic and heterographic homophones and, in a post hoc analysis, found nodifferences between the two types of homophone.


CUETOS ET AL.


We would have liked to have manipulated theage of acquisition (AoA) of the words as well(i.e., the age at which the words were learned),especially since many experiments have shownthat some of the effects attributed to frequencymay be due to AoA (e.g., Barry, Morrison, &Ellis, 1997; Cuetos, Ellis, & Alvarez, 1999;Morrison, Ellis, & Quinlan, 1992; for a review,see Johnston & Barry, 2006). However, given thescarcity of depictable homophones and the factthat high-frequency stimuli are also those thattend to be acquired early, it is very difficult tomanipulate these two variables. Therefore, weonly considered lexical frequency in the followingexperiments (although AoA is introduced as acovariate in the analyses of variance, ANOVAs).

Method

ParticipantsA total of 25 native speakers of Spanish (8 malesand 17 females with a mean age of 21.1 years),who were students at the School of Education ofthe University of Huelva, participated in thisexperiment. All had normal or corrected-to-normal vision.

StimuliA total of 95 simple black-and-white line drawingswere presented. The majority of the items weretaken from the Snodgrass and Vanderwart (1980)database. The other pictures were drawn by anartist in a style that was similar with regard tovisual complexity, thickness of line, and so on. Atotal of 38 of the pictures corresponded to 19homophones: 19 low- and 19 high-frequency

names. The other 38 pictures were controls: 19with names similar in frequency to the high-frequency homophones and 19 with names similarin frequency to the low-frequency homophones.The 19 remaining pictures were fillers. The fre-quencies of the stimuli were obtained from a two-million-word corpus (Alameda & Cuetos, 1995).The frequency of each homophone was estimatedspecifically for each sense by computing thenumber of times that each meaning appeared inthe corpus (e.g., for the word “banco”, we countedthe number of times that the word appeared withthe meaning signified by “bank” and the numberof times it appeared with the meaning “bench”).The stimuli are listed in Appendix A.

As can be seen in Table 1, the four conditionswere matched on name agreement and visual com-plexity (Fs, 1). The items were not matched onlength, although a majority of them had two orthree syllables. Name agreement and visual com-plexity scores were obtained from 30 studentswho had to name the pictures and rate the com-plexity of each drawing on a 7-point scale, where1 indicated “a very simple drawing” and 7 “a verycomplex drawing”.

ApparatusAPCwith an external voice keywas used. The soft-ware Superlab Pro (Abboud & Sugar, 1997) wasused for stimulus presentation and the collectionof data.

ProcedureParticipants were tested individually. Before thebeginning of the experiment itself, they wereasked to name the pictures in an untimed task,

Table 1. Summary of the statistical characteristics from Experiments 1a and 1b

HF LF

Homophones Controls Homophones Controls

Frequency 21.42 (23.68) 21.47 (23.83) 3.37 (3.82) 3.42 (3.96)Name agreement 90.28 (0.14) 93.08 (0.09) 93.14 (0.10) 94.14 (0.8)Visual complexity 2.65 (0.70) 2.98 (0.72) 2.80 (0.60) 3.11 (0.72)Length 4.74 (0.71) 5.89 (1.12) 4.74 (0.71) 5.79 (1.32)

Note: HF ! high frequency, LF ! low frequency. Standard deviations in parentheses.




and errors were corrected. The total number ofwords corrected was: 16 (3.37%) with the high-frequency homophones; 15 (3.16%) with thelow-frequency homophones; 14 (2.95%) with thehigh-frequency controls, and 21 (4.42%) withthe low-frequency controls. The screen wasplaced approximately 55 cm from the participants.In the experiment proper, the stimuli were pre-sented in a quasi-random order at the centre ofthe screen, preceded by four asterisks (!!!!). Theasterisks were presented for 1,500 ms and actedas a fixation point. The line drawings remainedon the screen until the participant initiated aresponse or 3,000 ms had elapsed. The partici-pants had to name each picture as fast as possiblewithout making errors. Latencies were recordedfrom picture onset and were monitored online byan experimenter who recorded any errors. Theexperiment started with 10 warm-up trials.

Results and discussion

Prior to the analysis of the naming latencies, weexcluded both errors and outliers (i.e., latenciesthat were three standard deviations above orbelow the mean) and trials in which the voicekey failed to record a response. A total of 5.5%of the responses were therefore discarded—3.24% corresponded to naming errors and2.26% to voice-key failures and outliers. Meannaming latencies and error rates are reportedin Table 2.

As shown in Table 2, high-frequency homo-phones were named faster than low-frequencyhomophones (76 ms). Moreover, naming latencies

for the low- and high-frequency homophoneswere very similar to those for their respectivefrequency-matched, nonhomophone controls. Ananalysis of variance with word type (homophonevs. nonhomophone) and frequency (high vs. lowfrequency) introduced as factors revealed nosignificant main effect of word type, F1 andF2 , 1, but a reliable effect of frequency, F1(1,24) ! 106.29, p , .001, MSE ! 147,478.27;F2(1, 18) ! 17.59, p , .01, MSE ! 106,552.02.The interaction between the two factors was notsignificant, F1 and F2 , 1. Since we had collectedAoA data for these stimuli (subjective ratings, ona 7-point scale, from 25 other students), AoAwas introduced as a covariate in a new analysis ofvariance. The AoA covariate was not significant(F , 1). Matched-pairs t tests were performed tofurther explore the nature of the frequency effect.High-frequency homophones were producedsignificantly faster than low-frequency homo-phones, t1(24) ! 8.71, p, .001; t2(18) ! 2.34,p , .05, and high-frequency controls were pro-duced significantly faster than low-frequencycontrols, t1(24) ! 5.62, p, .001; t2(18) ! 3.36,p , .01. However, no reliable difference wasobserved between low-frequency homophonesand low-frequency controls, t1 and t2 , 1, whilea significant difference was observed betweenlow-frequency homophones and high-frequencycontrols, t1(24) ! 10.18, p, .001; t2(18) ! 3.39,p , .01. That is, the naming times of the low-frequency homophones were similar to those ofthe low-frequency controls and different fromthose of the high-frequency controls.

The same pattern of results was found forerrors. The main effect of word type was notsignificant, Fs , 1. The main effect of frequencywas significant by participants only, F1(1,24) ! 4.38, p, .05, MSE ! 0.81; F2 , 1. Theinteraction between frequency and word type wasnot significant, Fs, 1.

These results are consistent with those reportedby Caramazza et al. (2001) and Bonin and Fayol(2002; at least with regard to the differencebetween high- and low-frequency homophonesin the latter study) and differ from those obtainedby Jescheniak and Levelt (1994). Low-frequency

Table 2. Mean naming latencies and percentage of errors as afunction of word frequency and homophony in Experiment 1a

HF LF

Latencies Errors Latencies Errors

Homophones 643 (78) 3.16 719 (79) 3.16Controls 638 (87) 2.53 715 (113) 4.42

Note: HF ! high frequency, LF ! low frequency. Latencies inms. Standard deviations in parentheses.


CUETOS ET AL.


homophones produced significantly longernaming times than did high-frequency homo-phones and high-frequency nonhomophone con-trols. Thus, picture naming latencies appear to bedetermined by specific-word frequency and nothomophone frequency. These results wereobtained in a language (Spanish) with a transpar-ent orthography like Dutch—the language usedin the original experiment by Jescheniak andLevelt. Thus, orthographic transparency cannotbe the reason for the contrasting results obtainedby Caramazza et al. and by Bonin and Fayol, andby Jescheniak and Levelt.

One possible explanation for the results inExperiment 1a is that the difference between thelow- and high-frequency homophones is due todifferences in the ease of recognition of the picturesused in the two sets of stimuli. It is possible that thepictures used for the low-frequency homophoneswere particularly hard to recognize, thus offsettingany advantage conferred by their homophonestatus. In other words, it is possible that thelonger naming times obtained with the low-fre-quency homophones were the result of the veryslow recognition times for the pictures used forthese stimuli compared to the recognition timesfor the high-frequency homophones and the non-homophone control stimuli. Although this seemsto be an unlikely explanation, we tested this possi-bility in a second experiment with the same stimuliin a task in which the participants did not have toname the objects but only to recognize them.

EXPERIMENT 1B: OBJECTDECISIONIN SPANISH

In this experiment, we used an object-decision tasksimilar to the one used by Kroll and Potter (1984).The participants had to decide whether or not astimulus was a familiar object. This task evaluatesthe cognitive operation necessary to pair perceivedvisual representations with stored structuralobject representations (Humphreys, Riddoch, &Quinlan, 1988) and can therefore serve as areasonable index of the ease with which an objectis recognized.

Method

ParticipantsA total of 25 native speakers of Spanish (9 malesand 16 females with a mean age of 21.0 years),taken from the same pool as that in the previousexperiment, participated in this experiment. Allhad normal or corrected-to-normal vision, andnone had taken part in the previous experiment.

StimuliThe stimuli were the same line drawings as thosethat were used in the previous experiment, alongwith 95 line drawings of nonobjects: 88 line draw-ings used by Kroll and Potter (1984) plus 7 newline drawings prepared specifically for this task.

ApparatusThe apparatus was the same as that in the previousexperiment, except that instead of a voice key, twoexternal buttons connected to the keyboard wereused to record responses.

ProcedureThe participants had to decide as fast as possiblewhether each given line drawing corresponded toa “real” or to an “unreal” object, and reactiontimes (RTs) were recorded. The “real” button wasassigned to the participant’s dominant hand, andthe “not-real” button to the nondominant hand.


Prior to the analysis of response times, errors wereremoved from the data. This resulted in theremoval of 7.9% of the responses: 2.4% forobjects and 5.5% for nonobjects. Of the data elimi-nated, 4.3% were trials on which participantsresponded erroneously, and the remaining 3.6%were outliers (three standard deviations above orbelow the mean).

The mean decision latencies and error rates inthe different experimental conditions are pre-sented in Table 3. The differences in RTsacross conditions were minimal (except in thecomparison with the response times for nonob-jects), and, importantly, the results indicate that




the low-frequency homophones were not moredifficult to recognize than the other stimuli. Asin Experiment 1a, ANOVAs were carried out fordecision latencies, with word type and frequencyas experimental factors. The main effect of wordtype (homophone vs. nonhomophone) was notsignificant, Fs, 1. High-frequency pictures tooklonger to categorize than low-frequency pictures.The main effect of frequency was significant byparticipants, F1(1, 24) ! 4.46, p, .05, MSE !3,641.40, but not by items, F2 , 1. The directionof this effect is opposite to the one found inExperiment 1a, where high-frequency words wereproduced faster than low-frequency words. Theinteraction between word type and frequencywas not significant, F1(1, 24) ! 1.69, MSE !1,309.28; F2 , 1. There were no reliable effectsfor errors, all Fs, 1.

To summarize, in object decision there were nosignificant differences between any of the exper-imental conditions. The only minor trend was forlow-frequency homophones, which were recog-nized faster than other stimuli—an effect thatruns counter to the one observed in the namingexperiment. Thus, the naming latency differencesfound between high- and low-frequency homo-phones and between low-frequency homophonesand high-frequency controls cannot be ascribed tothe object recognition stage but originate at somestage of lexical access.

EXPERIMENT 2A: PICTURENAMINGIN SPANISH

The findings of Experiments 1a and 1b are fullyconsistent with those reported by Caramazza

et al. (2001) in that the crucial variable indetermining naming latencies seems to be specific-word frequency and not homophone frequency.However, a number of potentially problematicproperties associated with the materials used inthese two experiments demand caution in inter-preting the results. In effect, it could be arguedthat the frequency differences between the high-and low-frequency members of the homophonepairs we used were not very large, thus perhapsmasking a real homophone effect. It should benoted, however, that we did obtain significantfrequency effects consistent with the predictionsof the specific-word frequency hypothesis, there-fore making it unlikely that the absence of ahomophone effect was due to the small differencesin frequencies between the high- and low-frequency members of the homophonous pairs.

Another potential problem concerns the fact thatthe experimental (homophone) and control (nonho-mophone) words were not perfectly matched onlength. However, length differences cannot beresponsible for the specific-word frequency effectsobtained with the homophone words since thesame phonological form was produced for thehigh- and low-frequency member of the homo-phone pair. Furthermore, the small differences inlength between the experimental and control itemsshouldhave favoured the faster production of homo-phones and, therefore, could not be responsible forthe differential effect of specific-word versus homo-phone frequency obtained in Experiment 1a.

Finally, there is the potential concern that theexperimental and control items were not matchedfor initial phoneme. Therefore, it remains possiblethat the differences observed for the initial pho-nemes of the picture names contributed to thepattern of results reported here.

As stated above, it is unlikely that the three poten-tially problematic factors mentioned above areresponsible for the pattern of results obtained inExperiment 1a, which suggests that it is specific-word and not homophone frequency that determinesthe speed of lexical access. Nonetheless, in orderto take account of these potential concerns, wereplicated Experiments 1a and 1b with a new setof stimuli.

Table 3. Mean object decision latencies and percentage of errors inExperiment 1b

HF LF





CUETOS ET AL.


Method

ParticipantsA total of 25 native speakers of Spanish (8 malesand 17 females with a mean age of 21.9 years),taken from the same pool as that in the previousexperiments, were recruited for this study. Allhad normal or corrected-to-normal vision, andnone had participated in any of the previousexperiments.

StimuliA total of 72 simple black-and-white drawingswere used. As in the previous experiment, amajority of the items were taken from theSnodgrass and Vanderwart (1980) database, anda few pictures were drawn by an artist in a stylethat was similar with regard to visual complexityand line thickness. A total of 36 of these drawingscorresponded to homophones: 18 representing themore frequent meaning, and 18 the less frequentmeaning. The other 36 drawings were controls:18 drawings whose names were similar in fre-quency to the high-frequency homophones, and18 similar in frequency to the low-frequencyhomophones. Homophones and control itemswere matched on length and initial phoneme.The frequency was obtained from a two-million-word corpus (Alameda & Cuetos, 1995). Nameagreement and visual complexity scores wereobtained from a separate group of 30 studentswho had to name the pictures and to rate thecomplexity of each drawing on a 7-point scale.

As shown in Table 4, the four conditionswere matched on name agreement, number of

phonemes, and visual complexity (Fs, 1). Thestimuli are listed in Appendix B.

ApparatusThe apparatus was identical to that used inExperiments 1a and 1b.

ProcedureThe procedurewas identical in all respects to that ofExperiment 1a. Before the beginning of the exper-iment, the participants were asked to name the pic-tures in an untimed task, and errors were corrected.The total number ofwords correctedwas: 20 (4.4%)in the high-frequency homophones; 30 (6.6%) inthe low-frequency homophones; 30 (6.6%) in thehigh-frequency controls; and 38 (8.4%) in thelow-frequency controls.


Prior to the analysis of the naming latencies, errorswere removed, as were outliers and other errorscaused by problems with the voice key. A total of9.8% of the trials were eliminated: 6.6% fornaming errors and the remainder (3.2%) as outliersand voice-key failures.

The mean naming latencies and percentages oferrors for the experimental and control conditionsare shown in Table 5.

High-frequency homophones were namedfaster (89 ms) than their low-frequency mates,and bothwere named as fast as their respective non-homophone frequency controls. There was noeffect of word type, F1 and F2 , 1, but a clearreliable effect of frequency, F1(1, 24) ! 43.09,

Table 4. Summary of the statistical characteristics from Experiments 2a and 2b

HF LF


Frequency 23.61 (22.19) 24.00 (21.83) 2.44 (2.85) 2.50 (2.41)Name agreement 96.11 (0.45) 94.8 (0.94) 93.33 (0.07) 94.81 (0.06)Visual complexity 2.39 (0.49) 2.60 (0.69) 2.72 (0.79) 2.76 (0.61)Length 4.89 (0.99) 4.67 (0.94) 4.89 (0.99) 5.22 (0.97)

Note: HF ! high frequency, LF ! low frequency. Standard deviations in parentheses.




p , .001, MSE ! 167,289.18; F2(1, 17) ! 6.65,p , .05, MSE ! 106,702.23. The interactionbetween the two factorswas not significant,Fs, 1.

Matched-pairs t tests show that high-frequencyhomophones were named significantly faster thanlow-frequency homophones, t1(24)! 5.94,p, .001; t2(17) ! 2.40, p, .05, and that high-fre-quency controls were named significantly faster thanlow-frequency controls, t1(24)! 5.27, p, .001;t2(17)! 2.05, p, .05. No reliable difference wasobserved between low-frequency homophones andlow-frequency nonhomophone controls, ts, 1.

As far as errors are concerned, the main effect offrequency was reliable, F1(1, 24) ! 9.99, p, .01,MSE ! 6.76; F2(1, 17) ! 12.17, p, .05, MSE !9.39, but the effect of word type was not, F1 andF2. The interaction between frequency and wordtype was also not significant, Fs, 1. Again an ageof acquisition covariate was included in the model,but its effect was not significant (F, 1).

The results obtained here fully replicate thoseobtained in Experiment 1a. However, beforediscussing these results further, it is important toconsider the possibility that these effects are dueto recognition differences. Experiment 2b wasdesigned to address this possibility by means of aword–picture matching task.

EXPERIMENT 2B: WORD–PICTUREMATCHING IN SPANISH

It is important to make sure that the picture-naming latency differences did not stem fromdifferences in the ease of recognizing the picturedobjects (Bonin, Chalard, Meot, & Barry, 2006).

In effect, we cannot reject the hypothesis thatlow-frequency homophones were named moreslowly than high-frequency homophones becauseof differences in the speed of recognizing the twosets of pictured objects (and not because of differ-ences in the speed with which the two sets of itemsare accessed from the mental lexicon). Of course,frequency may affect both picture recognitionand lexical access (but see Almeida, Knobel,Finkbeiner, & Caramazza, 2007; Bonin et al.,2006). As a result, it is not enough to comparenaming performance on the high- and low-frequency members of a homophone pair. Instead,we need to compare performance on these twotypes of words with that observed for appropriatelymatched nonhomophone control words. Thiscontrol condition makes it possible to assess theadditional effect, if any, of homophony in picturenaming. That is, the hypothesis that homophonesinherit the frequency of their homophone matesmakes two predictions: (a) Low-frequencymembers of homophone pairs should be namedfaster than specific-word frequency-matched con-trols (because homophones that have inherited thefrequency of their homophone mates should actu-ally be more frequent than the controls); and (b)low-frequency members of homophone pairsshould be named as fast as the high-frequencymembers of the pairs unless the latter are recog-nized faster, thereby leading to faster naminglatencies. It is therefore important to assesswhether the pictured objects in the various exper-imental conditions differ in the ease with whichthey are recognized.

In Experiment 1b, this possibility was assessedby asking participants to perform an objectdecision task. We found that the pictures of thelow-frequency homophone names were “recog-nized” as fast as those of their specific-wordfrequency controls and those of the high-frequencyhomophone names. This finding was taken asevidence that differences in picture recognitionspeed were not responsible for the naminglatency effects observed in the picture-namingtask. However, it could be argued that the objectrecognition task is not sufficiently sensitive toreveal subtle differences in the speed with which

Table 5. Mean naming latencies and percentage of errors as afunction of word frequency and homophone in Experiment 2a

HF LF





CUETOS ET AL.


pictures are recognized. To address this concern,we used a picture–word matching task inExperiment 2b. This task has been widely used inspeech production studies to control for nonlexicaldifferences between picture stimuli. Evidence vali-dating its use to control for difficulties arising at thelevel of nonlexical factors in picture naming hasrecently been provided by Stadhagen-Gonzalez,Damian, Perez, Bowers, and Marin (2009).

Method

ParticipantsA total of 25 native speakers of Spanish (7 malesand 18 females with a mean age of 21.9 years),again drawn from the same pool as in thoseinvolved in the previous experiments, participated.All had normal or corrected-to-normal vision, andnone had participated in any of the previousexperiments.

StimuliA total of 144 word–picture pairs were used. The72 pictures used in the previous experiment werepaired with semantically related words—forexample, the picture of a key (“llave”) was pairedwith the word “puerta” (door). A total of 72other pictures were paired with their correctname (e.g., the picture of a gorilla and the word“gorilla”) and were used as fillers. The reason forpresenting the targets in the negative trials wasto avoid word–object priming (and to use a differ-ent type of judgement from the one we used inExperiment 3b—see below).

ApparatusThe apparatus used in the previous experimentswas utilized in this experiment as well.

ProcedureA ready signal (!) was presented for 1,000 ms atthe centre of the screen. A word printed in bold(Times New Roman 50) was then presented for500 ms. Finally, a picture replaced the word andremained on the screen until the participantpressed one of two buttons connected to the

keyboard to indicate whether or not the wordwas the name of the picture.


The results are summarized in Table 6. The errorsproduced by the participants, 4.6% of the trials,were excluded from the RT analyses (2.8% weretrials on which participants had not pressed thecorrect key, and the remainder were outliers).

An ANOVA performed on the RTs failed tofind any significant differences between the differ-ent experimental conditions: Neither word type,F1 and F2 , 1, nor frequency, F1 and F2 , 1,had a reliable effect. The interaction betweenthese factors was also not significant, Fs, 1.Furthermore, no reliable effects were found forerrors, Fs, 1.

The results of this experiment confirm that thedifferences in naming latencies obtained inExperiment 2a cannot be attributed to differencesin the speed of picture recognition. Therefore, wefeel confident that the results of Experiment 2a, inwhich low-frequency homophones were namedmore slowly than high-frequency homophonesand no faster than specific-word frequency nonho-mophone controls, were due to the fact that it isspecific-word and not homophone frequency thatdetermines the speed of lexical access in speechproduction.

The findings obtained in Experiments 1a–1band 2a–2b are fully in line with those reported byCaramazza et al. (2001). They also partially repli-cate those reported by Bonin and Fayol (2002).However, as noted by both Jescheniak et al.

Table 6.Mean naming latencies and number of errors as a functionof word frequency and homophony in Experiment 2b

HF LF







(2003) and Biedermann and Nickels (2008a),Bonin and Fayol’s experiments did not include acontrol condition necessary to evaluate whetherit is specific-word or homophone frequency thatdetermines the speed of lexical access. Specifically,Bonin and Fayol (2002) compared naming timesbetween low- and high-frequency homophones(e.g., “ver”, worm, and “verre”, glass, both pro-nounced /vEr/) but did not include nonhomophonecontrols so that it could not be established whetherhomophony contributed to naming times. That is,even if it is true that low-frequency homophonesdo not fully inherit the frequency of their homo-phone mates (as found in Bonin & Fayol, 2002), itcould still be the case that homophony contributesto the speed of lexical access. This issue wasaddressed in the following experiment.

EXPERIMENT 3A: PICTURENAMINGIN FRENCH

Experiments 3a–3b were formally identical toExperiments 1a–1b and 2a–2b, except for twoaspects. First, the new experiments were inFrench and constituted a replication and, asnoted, a necessary extension of the experimentsreported by Bonin and Fayol (2002). Second, inExperiment 3a, participants were asked to namethe same stimuli multiple times, thereby allowingan assessment of the robustness of the frequencyeffect ( Jescheniak & Levelt, 1994).

Method

ParticipantsA total of 25 native speakers of French (6 malesand 19 females, with mean age 19.0 years), whowere students at Blaise Pascal University(Clermont-Ferrand, France), participated in theexperiment. All participants had normal or cor-rected-to-normal vision. Students received coursecredit for their participation.

StimuliThe stimuli consisted of 72 black-on-white draw-ings. Some were taken from various children’s

picture books, some were drawn by an artist, andsome were taken from the Alario and Ferrand(1999) database or from the Bonin, Peereman,Malardier, Meot, and Chalard (2003) database.We selected four sets of pictures. Two setsconsisted of 18 pairs of pictures with homophonicnames in French. All homophones were also het-erographic. The homophonic names for thesepicture targets were chosen such that one corre-sponded to a low-frequency (e.g., “ver”, worm)and one to a high-frequency reading of the homo-phone (e.g., “verre”, glass). These were the sameitems as those used in Bonin and Fayol’s (2002)study. The remaining 36 items were control pic-tures with nonhomophonic names. One set of 18pictures was the low-frequency, nonhomophonecontrol group and was matched on specific-wordfrequency with the low-frequency homophones;the other 18 pictures were the nonhomophonecontrols for cumulative-homophone frequency.Frequency values were taken from LEXIQUE(New, Pallier, Ferrand, & Matos, 2001) and fromMANULEX (Lete, Sprenger-Charolles, & Cole,2004). The LEXIQUE database is a corpus ofover 31 million words including sixteenth- totwentieth-century literacy texts, nineteenth- totwentieth-century scientific and technical texts,and regional variations; MANULEX provides fre-quency counts for words from a corpus of 1.9million words in the main French primary schoolreading books for four levels (1st, 2nd, 3rd, to5th grades) and for all grades. The picture namesare listed in Appendix C.

The picture targets were matched as closely aspossible on concreteness, name agreement, visualcomplexity, and word length (see Table 7).Name agreement and visual complexity scoreswere taken from Alario and Ferrand (1999) andfrom Bonin and Fayol (2002). Concretenessscores were taken from Bonin et al. (2003).

ApparatusThe experiment was performed with PsyScope,Version 1.2 (Cohen, MacWhinney, Flatt, &Provost, 1993), on a Macintosh computer. Thecomputer controlled the presentation of the picturesand recorded the latencies. The spoken latencies


CUETOS ET AL.


were recorded with a button box connected to thecomputer and an AIWA CM-T6 small tie-pinmicrophone connected to the button box.

ProcedureParticipants were tested individually. During apreliminary phase, participants had to revieweach picture and corresponding name. Sincename agreement scores were lower than those inthe previous experiments, a more stringent learn-ing phase was used. To this end, each picturewas presented on the screen with its namewritten below it, coupled with the auditory presen-tation of the name via headphones (SennheiserHD 25 SP). The picture remained on the screenuntil the participant pressed the space bar.Participants were told to look at each picture care-fully, to note its name and then, when they feltthey knew its name, to press the key to proceedto the next picture. Each trial had the followingstructure: A ready signal (!) was presented for1,000 ms, and the picture followed 200 ms later.The written and spoken names of the pictureswere presented 50 ms after picture onset. Whenthe participant pressed the space bar, the next trialbegan after a delay of 1,000 ms. To ensure thatparticipants had correctly learned the names associ-ated with the pictures, the experimenter tested

them on all pictures. The test was repeated on thepictures on which participants failed to providethe correct, intended name.2

The second phase was the experimental phase.Participants were told that they would see apicture (presented on the screen at a viewingdistance of approximately 60 cm), and they hadto name the picture as quickly as possible. Thefour sets of pictures were randomly presentedwithin a block. This block presentation wasrepeated four times, with a different randomiz-ation each time. The experimenter monitoredthe participants’ responses and scored them forcorrectness. The duration of the experiment didnot exceed one hour. A trial began with a readysignal (!), presented for 500 ms, and was followedby a picture. The picture remained on the screenuntil the participants initiated a spoken response.The next trial began 2,000 ms after the partici-pants had initiated their response. The sessionbegan with warm-up trials.


ANOVA was performed with frequency (highfrequency, low frequency), word type (homo-phone, nonhomophone), and presentation (first,second, third, fourth) as factors.

Table 7. Summary of the statistical characteristics from Experiment 3a

HF LF


Frantext freq. 123.7 (137.62) 116.8 (125.03) 13.5 (26.40) 14.4 (29.74)Manulex freq. 104.73 (82.88) 185.55 (156.32) 21.71 (25.47) 28.34 (49.71)Name agreement 83.06 (21.01) 76.28 (20.31) 66.39 (24.48) 70.78 (20.15)Visual complexity 2.44 (1.11) 2.69 (1.16) 3.05 (0.99) 2.88 (0.62)Length 4.44 (0.92) 5.17 (0.92) 4.39 (0.85) 5.06 (0.94)Concreteness 4.07 (0.74) 4.24 (0.56) 4.31 (0.58) 4.52 (0.35)

Note:HF ! high frequency, LF ! low frequency; Frantext freq. ! objective word frequency taken from Frantext (LEXIQUE: Newet al., 2001); Manulex freq. ! Child frequency taken from Lete et al. (2004). Standard deviations in parentheses.

2 Learning times were not recorded in this experiment. However, in the Bonin and Fayol (2002) study, in which learning timeswere recorded, the analyses revealed that they had no impact on the picture-naming performance. Meyer and Damian (2007) haveinvestigated the impact of the familiarization phase on picture-naming performance. Using the picture–picture interference para-digm, they failed to find an effect of the practice phase on naming times.




Observations were discarded from the latencyanalyses if a participant did not respond withinthe allotted time window, a technical problemoccurred, or an item other than the expected onewas produced. Observations were also scored aserrorswhen participants stuttered or produced non-linguistic sounds (such as mouth clicks) or repairedthe utterance after a dysfluency. Moreover,latencies exceeding two standard deviations abovethe participant and item means were discarded(2.11% of the data). Overall, 7.85% of the obser-vations were discarded from the latency analyses.

As in the previous experiments, the analyseswere performed on both latencies and error rates.Only the significant results of the latter analysesare reported.

Mean spoken latencies and error rates for eachset of pictures and at each presentation level areshown in Table 8.

LatenciesLatencies were shorter with the repeated presen-tation of the items, F1(3, 72) ! 87.33,MSE ! 6,393.33, p , .001; F2(3, 204) ! 254.36,MSE ! 1,598.26, p , .001, and shorter on high-frequency items than on low-frequency ones,F1(1, 24) ! 78.95, MSE ! 2,760.40, p , .001;F2(1, 68) ! 8.13, MSE ! 20,616.73, p , .01.The main effect of word type was significant byparticipants, F1(1, 24) ! 5.65, MSE ! 3,202.86,p , .05, but not in item-wise analyses, F2 , 1.The size of the frequency effect decreased withrepeated picture presentation. This interactionwas significant in both the by-participant andby-item analyses, F1(3, 72) ! 7.36, MSE !1,494.73, p , .001; F2(3, 204) ! 4.65, MSE !1,598.26, p , .005. Word type did not interactsignificantly with either presentation, F1(3,72) ! 1.30, MSE ! 1,119.18; F2 , 1, or fre-quency, Fs, 1. Finally, the size of the wordfrequency effect did not vary reliably as a function

of either word type or presentation, F1(3,72) ! 1.04, MSE ! 1,441.92; F2 , 1. Plannedcomparisons (all ps, .05) revealed that high-frequency homophonic pictures were namedfaster than low-frequency homophonic picturesat each level of presentation in both the subject-wise and item-wise analyses. Moreover, high-frequency nonhomophonic pictures producedfaster responses than low-frequency nonhomo-phonic pictures, at each level of presentation.Except for the third presentation (and in thesubject-wise analysis), high-frequency homo-phones did not produce faster responses thantheir high-frequency controls. Low-frequencyhomophones were not named faster than theircontrols (except for the second presentation inboth the subject-wise and item-wise analyses).Finally, contrary to the expectations derived fromJescheniak and Levelt’s (1994) homophone fre-quency inheritance hypothesis, low-frequencyhomophones produced slower responses thantheir matched cumulative frequency controls ateach level of presentation.3

ErrorsThe number of errors decreased over repeatedpresentation of the items, F1(3, 72) ! 33.95,MSE ! 14.99, p, .001; F2(3, 204) ! 25.69,MSE ! 14.26, p, .001. Also, there were fewererrors on high- than on low-frequency items,F1(1, 24) ! 16.53, MSE ! 17.37, p, .001; F2(1,68) ! 4.82,MSE ! 42.89, p, .05. The interactionbetween presentation and frequency was signifi-cant, F1(3, 72) ! 5.93, MSE ! 10.29, p, .01;F2(3, 204) ! 3.08, MSE ! 14.26, p, .05, andthe interaction between word type and frequencywas significant in the subject-wise analysis only,F1(1, 24) ! 5.91, MSE ! 17.86, p, .05; F2(1,68) ! 1.77, MSE ! 42.89.

Before interpreting these results further, it isimportant to establish that they could not have

3 It could be argued that the name agreement scores distribution is somewhat reflected in the naming latencies. However, whenthese scores were introduced as a covariate in the by-item analyses, the pattern of results remained the same. Moreover, and impor-tantly, low-frequency homophones were named more slowly than their matched cumulative frequency controls with name agreementscores controlled for: The name agreement scores did not reliably differ between high-frequency controls (76%) and low-frequencyhomophones (66%).


CUETOS ET AL.


arisen as a consequence of differences in the speedof recognition of the pictures used for the low- andhigh-frequency homophones and their controls.We address this issue in Experiment 3b.

EXPERIMENT 3B: WORD–PICTUREMATCHING IN FRENCH

Method

ParticipantsA total of 25 psychology students (5 males and 20females, with a mean age of 18 years), drawn fromthe same pool as those who participated inExperiment 3a, participated in this experimentand received a course credit.

ProcedureParticipants were tested individually. Participantswere told that they would see a printed word fol-lowed by a picture, and they were instructed todecide as quickly as possible whether the worddenoted the same concept as the picture. A “go/no-go” task was used instead of a “same/different”task. Participants responded by pressing a responsebutton (“go” response) when the name and thepicture referred to the same concept and madeno response otherwise (“no-go” response). Thetype of task was chosen in order to contrast withthe task used in Experiment 2b (same/different)and thus broaden the range of measures used to

assess the potential contribution of picture recog-nition to participants’ naming performance.

Each trial began with a warning signal (!),presented for 500 ms, which was followed by aprinted word presented for 1,000 ms. After adelay of 1,000 ms, a picture was presented. In allcases, the picture disappeared after a delay of2,000 ms. An intertrial interval of 2 s followed.In contrast to Experiment 3a, the pictures werepresented only once. The session began with awarm-up trial and lasted approximately half anhour.

Stimuli and apparatusWe selected an additional set of 72 pictures withthe same characteristics as the target pictures inorder to create the no-go trials. The apparatuswas the same as that in Experiment 3a.

Trials on which participants responded erro-neously were discarded, and RTs exceeding twostandard deviations above the participant anditem means were excluded (2.17%). Overall,4.89% of all trials were discarded. The analyseswere performed on both RTs and error rates (onlythe significant results of the latter analyses arereported).


Mean RTs and error rates (in percentages) areshown in Table 9.

Table 8. Mean naming latencies and error rates as a function of word frequency, homophony, and presentation in Experiment 3a

HF LF


Presentation Latencies Error rates Latencies Error rates Latencies Error rates Latencies Error rates

1 925 (103) 4.0 931 (129) 4.0 1007 (97) 9.6 995 (145) 5.82 839 (97) 0.2 837 (96) 0.7 904 (113) 3.11 875 (119) 2.23 809 (72) 0.4 787 (79) 1.3 844 (83) 1.55 824 (72) 1.34 793 (74) 0.7 777 (81) 1.6 816 (66) 2.0 805 (65) 0.9Overall 842 (100) 1.3 833 (114) 1.9 893 (116) 4.1 875 (128) 2.6

Note: HF ! high frequency, LF ! low frequency. Naming latencies in ms. Standard deviations in parentheses. Error rates inpercentages.




Reaction timesThe main effect of frequency was not significant,Fs, 1, but the main effect of word type was, F1(1,24)! 38.76, MSE! 2,658.33, p, .001; F2(1,68)! 8.63, MSE! 9,828.20, p, .01. The inter-action between word type and frequency was notreliable,F1(1, 24) ! 1.14,MSE ! 2,433.93;F2 , 1.

ErrorsThere were more errors on low- than on high-frequency items. However, the main effect offrequency was significant only in the subject-wiseanalysis, F1(1, 24)! 5.68, MSE! 0.40, p, .05;F2 , 1. The interaction between word type andfrequency was reliable, F1(1, 24) ! 12.69, p, .01;MSE ! 0.35; F2(1, 68) ! 4.03, MSE ! 1.52,p , .05.

The pattern of results obtained in the name–object verification task did not mirror the patternof results found in the picture-naming task.More specifically, we did not find that the picturesof low-frequency homophones were recognizedwith substantially greater difficulty than those ofhigh-frequency homophones or those of nonho-mophone controls. Thus, the word frequencyeffects observed in Experiment 3a cannot beattributed to differences arising at the level ofobject identification. Moreover, additional ana-lyses performed on the naming times inExperiment 3a, with object verification times

introduced as covariates, did not alter the patternof findings obtained in the picture-naming task.4

GENERAL DISCUSSION

In three picture-naming experiments and theirassociated control experiments we have confirmedthat the frequency that determines the speed oflexical access in the production of homophones isspecific-word frequency and not homophonefrequency. That is, when producing the words“vela” (the Spanish word for “candle”), “ver” (theFrench word for “worm”), or “nun”, what mattersis the specific frequency of the word and not thesum of their individual frequencies and those oftheir respective homophones (e.g., the frequencyof “nun” plus “none”). Across three experiments,we found that in lexical access for speechproduction, the naming of homophones such as/bela/ (candle) and /ver/ (worm) does notbenefit from the existence of other words thathave the same pronunciation in Spanish andFrench, respectively. That is to say, there arereliable differences between the production of thehigh- and the low-frequency uses of homophones(e.g., naming the pictures of a candle and a sail,both of which are pronounced /bela/ in Spanishor the pictures of a glass and a worm, both ofwhich are pronounced /ver/ in French), and the

Table 9. Mean verification times and error rates as a function of word frequency andhomophony in Experiment 3b

Latencies Error rates

HF Homophones 694 (113) 1.56Controls 641 (94) 2.22

LF Homophones 711 (112) 5.56Controls 636 (101) 1.56

Note: HF ! high frequency, LF ! low frequency. Naming latencies in ms. Standarddeviations in parentheses. Error rates in percentages.

4 We have tested whether frequency trajectories (which according to Bonin, Barry, Meot, & Chalard, 2004, and Zevin &Seidenberg, 2002, is an objective operationalization of the age/order of acquisition of the words) varied as a function of homophonyand word frequency for the items used in the French experiments. Frequency trajectory norms did not vary reliably on either dimen-sion. Therefore, on this reasoning, the word frequency effects found in picture-naming latencies in Experiment 3a cannot be attrib-uted to AoA/order of acquisition.


CUETOS ET AL.


production latencies of homophones are equivalentto those of other words of similar frequencythat are not homophones. These results wereobtained in the context of experiments in whichwe controlled for various factors—name agree-ment, visual complexity, and phonological com-plexity—that could have contributed to variationin naming latencies.

The picture-naming results cannot be attributedto processes other than lexical ones, since the resultsof the object decision and object–name matchingtasks rule out an explanation in terms of differencesin picture recognition. Nor can the differences beattributed to postlexical processes, since the relevantcontrast involved the productionof the samephono-logical forms (homophones). Therefore, the differ-ences observed between high- and low-frequencyhomophones can arise only at some stage of lexicalaccess (see Knobel, Finkbeiner, & Caramazza,2008, for an extensive discussion of the locus oflexical frequency in speech production).

Finally, the results confirm that it is specific-word frequency of each sense of a homophoneand not homophone frequency that determinesnaming latencies. This conclusion is supportedby the observation that low-frequency homo-phones are produced significantly more slowlythan nonhomophone controls matched on fre-quency with the cumulative frequency of thehomophones and that they are produced nofaster than nonhomophone controls matched onfrequency with the frequency of the specific senseof the homophones that were tested.

The results we report directly address the con-cerns expressed by Jescheniak et al. (2003) aboutthe interpretation of previous research intospecific-word and homophone frequency effects.First, it is now clear that differences in ortho-graphic transparency across languages are not thecause of contrasting findings concerning the rolesof specific-word and homophone frequency inlexical access to homophones. More specifically,we have found that it is specific-word and nothomophone frequency that determines lexicalaccess times in a language with transparent ortho-graphy (Spanish), thus replicating similar results inlanguages with opaque orthographies (English and

Mandarin Chinese; Caramazza et al., 2001).Second, the specific-word frequency effect isobtained when controlling for possible differencesin the ease of recognition across stimulus sets usinga variety of control tasks (object decision, “no”responses in picture–word matching, and “go”responses in “go/no-go” picture–word matching)that are generally assumed to reveal effectsoccurring at the level of object recognition.

Third, because we used nonhomophone fre-quency-matched controls for both the low- andhigh-frequency uses of a homophone we can defi-nitely conclude that the effective frequency indetermining naming latencies is the frequency ofeach individual sense of a homophone, and there-fore that homophones do not inherit the frequencyof their homophone mates. And, finally, becausewe replicated the specific-word frequency effectin three experiments with different stimuli intwo different languages, we can be sure that theeffect is highly reliable. Indeed the empiricalgeneralization that it is specific-word and nothomophone frequency that determines the speedof lexical access has been shown to hold acrossfour languages that vary considerably inorthographic transparency (Spanish . French .English . Chinese) and morphophonologicalcomplexity (Chinese. English . French .Spanish). This gives us considerable confidencethat the phenomenon is a general property oflexical access in language production. As such itrepresents a crucial phenomenon that must beaccounted for by theories of lexical access.

What are the implications of the specific-wordfrequency effect for theories of lexical access? Oneclear implication is that it is specifically thefrequency of each individual meaning thatdetermines the ease of lexical access in languageproduction. This implies that the frequencyeffect has its locus at the level at which the differ-ent members of a homophone set are distinguishedfrom each other, and a word’s grammatical proper-ties (grammatical class, grammatical gender, etc.)are specified (Finocchiaro & Caramazza, 2006).This conclusion is consistent with at least twotypes of model of the lexicon. One type of modelconsists of those models that assume (a) that




there is only a single lexical layer mediatingbetween a word’s meaning and its phonologicalcontent and (b) that the members of a homophoneset are represented as fully independent, frequency-sensitive lexical nodes (e.g., Caramazza, 1997; seeFigure 1b). The other type of model consists ofthose that assume two lexical layers but locate thefrequency effect at the lemma level where themembers of a homophone set are distinguished(e.g., Dell, 1990). Our findings are not consistentwith those two-layer models that (a) assume thathomophones share a lexical representation (thelexeme level) and (b) locate the frequency effect atthat level of lexical access ( Jescheniak & Levelt,1994; Levelt, Roelofs, & Meyer, 1999).

Original manuscript received 26 January 2009Accepted revision received 22 April 2009First published online 2 September 2009

REFERENCES

Abboud, H., & Sugar, D. (1997). SuperLab Pro:Experimental Laboratory software [Computersoftware]. Phoenix, AZ: Cedros.

Alameda, J. R., & Cuetos, F. (1995). Diccionario de fre-cuencias de las unidades linguısticas del castellano[Frequency dictionary of Spanish linguistic units].Oviedo, Spain: Servicio de Publicaciones de laUniversidad de Oviedo.

Alario, X., & Ferrand, L. (1999). A set of 400 picturesstandardized for French: Norms for nameagreement, image agreement, familiarity, visualcomplexity, image variability, and age of acquisition.Behavior Research Methods, Instruments, &Computers, 31, 531–552.

Almeida, J., Knobel, M., Finkbeiner, M., & Caramazza,A. (2007). The locus of frequency effect in picturenaming: When recognizing is not enough.Psychonomic Bulletin & Review, 14, 1177–1182.

Barry, C., Morrison, C. M., & Ellis, A. W. (1997).Naming the Snodgrass and Vanderwart pictures:Effects of age of acquisition, frequency and nameagreement. Quarterly Journal of ExperimentalPsychology, 50A, 560–585.

Biedermann, B., Blanken, G., & Nickels, L. A. (2002).The representation of homophones: Evidence fromremediation. Aphasiology, 16, 935–980.

Biedermann,B.,&Nickels,L. (2008a).Homographic andheterographic homophones in speech production.Does orthography matter? Cortex, 44, 683–697.

Biedermann, B., & Nickels, L. (2008b). The represen-tation of homophones: More evidence from theremediation of anomia. Cortex, 44, 276–293.

Bonin, P., Barry, C., Meot, A., & Chalard, M. (2004).The influence of age of acquisition in word readingand other tasks: A never ending story? Journal ofMemory and Language, 50, 456–476.

Bonin, P., Chalard, M., Meot, A., & Barry, C. (2006).Are age-of-acquisition effects on object namingsimply due to differences in object recognition?Comments on Levelt (2002). Memory & Cognition,34, 1172–1182.

Bonin, P., & Fayol, M. (2002). Frequency effects in thewritten and spoken production of homophonicpicture names. European Journal of CognitivePsychology, 14, 289–313.

Bonin, P., Meot, A., Aubert, L., Malardier, N.,Niedenthal, P., & Capelle-Toczek, M.-C. (2003).Normes de concretude, de valeur d’imagerie, defrequence subjective et de valence emotionnellepour 867 mots [Concreteness, imageability,subjective frequency and emotional valence normsfor 867 words]. L’Annee Psychologique, 104, 655–694.

Bonin, P., Peereman, R., Malardier, N., Meot, A., &Chalard, M. (2003). A new set of 299 pictures forpsycholinguistic studies: French norms for nameagreement, image agreement, conceptual familiarity,visual complexity, image variability, age of acqui-sition and name latencies. Behavior ResearchMethod, Instruments & Computers, 35, 158–167.

Caramazza,A. (1997).Howmany levels of processing arethere in lexical access? Cognitive Neuropsychology, 14,177–208.

Caramazza, A., Bi, Y, Costa, A., & Miozzo, M. (2004).What determines the speed of lexical access:Homophone or specific-word frequency? A reply toJescheniak et al. (2003). Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 30,278–282.

Caramazza, A., Costa, A., Miozzo, M., & Bi, Y. (2001).The specific-word frequency effect: Implicationsfor the representation of homophones in speechproduction. Journal of Experimental Psychology:Learning, Memory, and Cognition, 27, 1430–1450.

Cohen, J., McWhinney, B., Flatt, M., & Provost, J.(1993). PsyScope: An interactive graphic systemfor designing and controlling experiments in thepsychology laboratory using Macintosh computers.


CUETOS ET AL.


Behavior Research Methods, Instruments, & Computers,25, 257–271.

Cuetos, F., Ellis, A. W., & Alvarez, B. (1999). Namingtimes for the Snodgrass and Vanderwart pictures inSpanish. Behavior Research Methods, Instrumentsand Computers, 31, 650–658.

Cutting, J. C., & Ferreira, V. S. (1999). Semantic andphonological information flow in the productionlexicon. Journal of Experimental Psychology:Learning, Memory, and Cognition, 25, 318–344.

Dell, G. (1990). Effects of frequency and vocabularytype on phonological speech errors. Language andCognitive Processes, 5, 313–349.

Ferreira, V. S., & Griffin, Z. M. (2003). Phonologicalinfluences on lexical (MIS)selection. PsychologicalScience, 14, 86–90.

Finocchiaro, C., & Caramazza, A. (2006). Theproduction of pronominal clitics: Implications fortheories of lexical access. Language and CognitiveProcesses, 21(1–3), 141–180.

Gahl, S. (2008). Time and thyme are not homophones:The effect of lemma frequency on word durations inspontaneous speech. Language, 84, 474–496.

Humphreys, G. W., Riddoch, M. J., & Quinlan, P. T.(1988). Cascade processes in picture identification.Cognitive Neuropsychology, 5, 67–103.

Jescheniak, J. D., & Levelt, W. J. M. (1994). Wordfrequency effects in speech production: Retrievalof syntactic information and of phonologicalform. Journal of Experimental Psychology: Learning,Memory, and Cognition, 20, 824–843.

Jescheniak, J. D., Meyer, A. S., & Levelt, W. J. M.(2003). Specific-word frequency is not all thatcounts in speech production: Comments onCaramazza, Costa, et al. (2001) and new experimen-tal data. Journal of Experimental Psychology: Learning,Memory, and Cognition, 29, 432–438.

Johnston, R. A., & Barry, C. (2006). Age of acquisitionand lexical processing. Visual Cognition, 13,789–845.

Knobel, M., Finkbeiner, M., & Caramazza, A. (2008).The many places of frequency: Evidence for a novellocus of the lexical frequency effect in wordproduction. Cognitive Neuropsychology, 25, 256–286.

Kroll, J. F., & Potter, M. C. (1984). Recognizing words,pictures and concepts: A comparison of lexical, object

and reality decisions. Journal of Verbal Learning andVerbal Behaviour, 23, 39–66.

Lete, B., Sprenger-Charolles, L., & Cole, P. (2004).MANULEX: A grade-level lexical database fromFrench elementary-school readers. Behavior ResearchMethods, Instruments, & Computers, 36, 156–166.

Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). Atheory of lexical access in speech production. Brainand Behavioral Sciences, 22, 1–75.

Meyer, A. S., & Damian, M. F. (2007). Activation ofdistractor names in the picture-picture interferenceparadigm. Memory and Cognition, 35, 494–503.

Miozzo, M., & Caramazza, A. (2005). The represen-tation of homophones: Evidence from the distractorfrequency effect. Journal of Experimental Psychology:Language, Memory, and Cognition, 31, 1360–1371.

Miozzo, M., Jacobs, M., & Singer, N. (2004). Therepresentation of homophones: Evidence fromanomia. Cognitive Neuropsychology, 21, 840–866.

Morrison, C. M., Ellis, A.W., & Quinlan, P. T. (1992).Age of acquisition, not word frequency, affects objectnaming, not object recognition. Memory andCognition, 20, 705–714.

New, B., Pallier, C., Ferrand, L., & Matos, R. (2001).Une base de donnees lexicales du francaiscontemporain sur Internet: LEXIQUE [A lexicaldatabase of contemporary French on Internet:LEXIQUE]. L’Annee Psychologique, 101, 447–462.

Shatzman, K. B., & Schiller, N. O. (2004). The wordfrequency effect in picture naming: Contrastingtwo hypotheses using homonym pictures. Brain andLanguage, 90, 160–169.

Snodgrass, J. G., & Vanderwart, M. (1980). A standar-dized set of 260 pictures: Norms for name agree-ment, image agreement, familiarity, and visualcomplexity. Journal of Experimental Psychology:Human Learning and Memory, 6, 174–215.

Stadhagen-Gonzalez, H., Damian, M. F., Perez, M. A.,Bowers, J. S., & Marin, J. (2009). Name-pictureverification as control measure for object naming:A task analysis and norms for a large set of pictures.Quarterly Journal of Experimental Psychology, 62,1581–1597.

Zevin, J. D., & Seidenberg, M. S. (2002). Age of acqui-sition effects in word reading and other tasks. Journalof Memory and Language, 47, 1–29.




APPENDIX A

Stimuli used in Experiments 1a and 1b

High frequency Low frequency


banco bank chaqueta banco bench cunabota boot tarta bota wine skin mofetabote tin can martillo bote boat focacarta letter policıa carta playing-card aguacatecometa comet mariposa cometa kite ciruelagato cat arbol gato jack almejaglobo globe cerilla globo balloon cascabelladron thief mascara ladron multiple plug fuellellama flame navaja llama llama guadanametro meter huevo metro subway leonmuelle quay jaula muelle spring dianamuneca wrist aguila muneca doll grifonuez walnut pina nuez Adam’s apple langostapala shovel estufa pala power shovel flexopico woodpecker antorcha pico pickaxe cangrejoplanta plant anillo planta foot arponpluma pen piscina pluma feather tenedorradio radio iglesia radio radius lupavela candle monja vela sail tiburon

APPENDIX B




bota boot buho bota wine skin buzobote tin can baul bote boat buzoncarta letter cafe carta playing-card cebracinta strip cesta cinta video cojınescalera stairs estrella escalera steps extintorfalda skirt barba falda foothill farolagato cat gafas gato jack gongglobo globe grua globo balloon gaitaladron thief lavabo ladron multiple plug lombrizllama flame leon llama llama llantallave key saco llave monkey wrench lincemetro metre moneda metro subway monja

(Continued overleaf )


CUETOS ET AL.


APPENDIX C




encre ink corde rope ancre anchor crepe pancakeballe ball doigt finger bal dance rat ratcanne cane jupe skirt cane duck cintre hangerchaıne chain regle ruler chene oak bombe bombcoeur heart fille girl choeur chorus coffre treasure chestcol collar cercle circle colle glue paume palmcoq rooster masque mask coque hull brique brickcorps body femme woman cor horn pion pawnsigne sign lettre letter cygne swan puzzle jigsaw-puzzlelait milk lune moon laie wild sow chope beer muglac lake lance spear laque hair spray urne urnlutte wrestling croix cross luth lute poulpe octopusmur wall langue tongue mure blackberry sphinx sphinxpoint point porte door poing fist banc benchpoids weight nez nose pois pea douche showerreine queen feuille leaf renne reindeer tank tanksang blood table table cent hundred piece coinverre glass groupe group ver worm biche deer

Appendix B (Continued)



pico woodpecker pino pico pickaxe pomoplanta plant placa planta sole panalpluma pen preso pluma feather posteradio radio reloj radio radius roperoraton mouse rosa raton mouse (PC) riflevela candle vaca vela sail venda




The speciﬁc-word frequency effect in speech production...

Documents

Transcript of The speciﬁc-word frequency effect in speech production...